a 1-d radial vaneless diffuser model accounting for the ... · a 1-d radial vaneless diffuser model...
TRANSCRIPT
A 1-D radial vaneless diffuser model accounting for the effects ofaerodynamic blockage
Stuart, C., Spence, S., Kim, S., Filsinger, D., & Starke, A. (2015). A 1-D radial vaneless diffuser modelaccounting for the effects of aerodynamic blockage. 1-12. Paper presented at International TurbochargingSeminar 2015, Tianjin, China.
Document Version:Peer reviewed version
Queen's University Belfast - Research Portal:Link to publication record in Queen's University Belfast Research Portal
Publisher rightsCopyright 2015 The Authors
General rightsCopyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or othercopyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associatedwith these rights.
Take down policyThe Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made toensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in theResearch Portal that you believe breaches copyright or violates any law, please contact [email protected].
Download date:13. Jul. 2018
1
Abstract The radial vaneless diffuser, though comparatively simple in
terms of geometry, poses a significant challenge in obtaining
an accurate 1-D based performance prediction due to the
swirling, unsteady and distorted nature of the flow field.
Turbocharger compressors specifically, with the ever
increasing focus on achieving a wide operating range, have
been recognised to operate with significant regions of
spanwise separated flow, particularly at off-design
conditions.
Using a combination of single passage Computational Fluid
Dynamics (CFD) simulations and extensive gas stand test
data for three geometries, the current study aims to evaluate
the onset and impact of spanwise aerodynamic blockage in
radial vaneless diffusers, and how the extent of the blocked
region throughout the diffuser varies with both geometry
and operating condition. Having analysed the governing
performance parameters and flow phenomena, a novel 1-D
modelling method is presented and compared to an existing
baseline method as well as test data to quantify the
improvement in prediction accuracy achieved.
Nomenclature A Flow area (m
2)
AR Area ratio of diffuser (-)
b Passage height (m)
B Blockage (-)
Cf Skin friction coefficient (-)
CP Static pressure recovery coefficient (-)
D Diameter (m)
I Ideal / Isentropic
k Skin friction constant (-)
ṁ Mass flow rate (kg/s)
p Static pressure (Pa)
p0 Total pressure (Pa)
PR Total-total pressure ratio (-)
r Radius (m)
R Gas constant (J/kgK)
Re Reynolds number (-)
RR Radius Ratio of diffuser (-)
T Temperature (K)
T0 Total temperature (K)
U Blade speed (m/s)
V Absolute velocity (m/s)
V Volumetric flow rate (m3/s)
YD Diffuser loss coefficient (-)
γ Ratio of specific heats (-)
α2 Mean impeller tip flow angle relative to radial (°)
ϕ Local flow coefficient (-)
η Isentropic total-total efficiency (-)
ρ Density (kg/m3)
ϟ Diffuser inlet flow parameter (-)[ϟ =��√𝛾𝑅𝑇01
𝑈22𝐷2
2 ]
β Flow angle relative to meridional (deg)
1-D One-dimensional
SFM Swirl flow meter
VLD Vaneless diffuser
Subscripts:
i Calculation step
I Ideal
r Radial direction
TT Total –to-total
u Tangential direction
1 Stage inlet
2 Impeller exit / vaneless diffuser inlet
3 Vaneless diffuser exit
4 Volute exit measurement plane
max Maximum value for a given parameter
1. Introduction The radial vaneless diffuser, though comparatively simple in
terms of geometry, poses a significant challenge in obtaining
an accurate 1-D performance prediction due to the highly
swirling, distorted and unsteady nature of the flow field
emanating from the impeller. The distorted nature of the
flow has constituents in both the pitchwise and spanwise
directions, arising from the characteristic jet-wake impeller
exit flow field and the curvature of the shroud wall
respectively. Clearly, accounting for these flow features is
of paramount importance if a clear indication of
performance is to be obtained, particularly at off-design
conditions.
A recent focus on the characterisation of impeller
recirculation by Harley et al. [1] demonstrated the
propensity of modern automotive turbochargers to operate
with significant regions of recirculation at the inlet, leading
to, among other impacts, significant regions of aerodynamic
blockage being presented to the incoming flow. Impeller
exit recirculation too has received some attention in recent
years, with Qiu et al. [2] providing a means of
A 1-D RADIAL VANELESS DIFFUSER MODEL ACCOUNTING FOR THE
EFFECTS OF AERODYNAMIC BLOCKAGE
C. Stuart1*
, S.W. Spence1, S. Kim
1, D. Filsinger
2 & A. Starke
2
1 School of Mechanical and Aerospace Engineering, Queen’s University Belfast, UK
2 IHI Charging Systems International GmbH, Germany
*corresponding author: Tel.:+4428-9097-4569; fax: +4428-9097-4178
E-mail address:[email protected]
2
characterising the recirculation present in the blade to blade
plane. As is frequently the case with centrifugal compressor
aerodynamics, research relating to the impeller has led the
way. The current work however aims to redress the balance
and evaluate the impact of flow separation and recirculation
and the associated spanwise aerodynamic blockage provided
to the flow in radial vaneless diffusers.
Previous work by the authors [3] relating to an evaluation
of existing 1-D vaneless diffuser modelling methods for
turbocharger centrifugal compressor applications
highlighted the importance of accounting for the presence of
spanwise aerodynamic blockage at the entrance to the
diffuser when specifying the flow field. An approach
involving the use of single passage CFD simulations for
each of the compressor stages in question was utilised to
permit a correlation capturing the influence of geometry and
operating condition to be generated. A new diffuser inlet
flow parameter, as depicted in Eq. (1), also had to be
specified in order to allow the variation of diffuser inlet
blockage with both geometry and operating condition to be
captured within a representative non-dimensional parameter
that could be applied to any data set. The resulting flow
parameter, which captures the dominant diffuser inlet flow
variables identified by Cumpsty [4] equates to the quotient
of local flow coefficient and impeller tip Mach number,
henceforth represented by ϟ.
ϟ =��√𝛾𝑅𝑇01
𝑈22𝐷2
2 (1)
The resulting empirical relationship between diffuser inlet
blockage B2 and the diffuser inlet flow parameter ϟ
amounted to a quartic fit line, as shown by Eq. (2).
𝐵2 = 91900ϟ4 − 62100ϟ3 + 15500ϟ2 − 1730ϟ + 90.6
(2)
The characterisation work pertaining to diffuser inlet
aerodynamic blockage demonstrated the presence of
significant levels of blockage (up to 60%), particularly
towards the surge side of the map. Specification of this
parameter permitted the complete flow field at inlet to the
diffuser to be calculated from gas stand test data gathered at
QUB, allowing an evaluation of different existing meanline
diffuser modelling methods to be undertaken.
The resulting modelling evaluation highlighted that the use
of an equivalent skin friction coefficient (Cf) as a bulk loss
term in 1-D diffuser modelling can, when tuned correctly,
deliver a performance prediction within an acceptable
window of accuracy. However, such a method does not
permit the designer to interrogate the results from the model
and easily identify the predominant sources of loss, making
it of limited use as a design tool.
The current work focuses on quantifying the degree of
spanwise separation of the flow, and how this is related to
geometry and operating condition. While it would be ideal
to directly evaluate the extent of the resulting aerodynamic
blockage presented to the flow from test data, the scale of
the stages being investigated does not permit the necessary
instrumentation to be reasonably incorporated. The
alternative approach applied for the current study was the
use of single passage CFD simulations for each geometry;
the validation of each against gas stand test data is presented
in a subsequent section.
Comparisons based on the resulting performance prediction
were drawn between the proposed modelling method, the
baseline Herbert [5] vaneless diffuser model and gas stand
test data. Three modern automotive turbocharger centrifugal
compressors, denoted C-4, C-5 and C-6 respectively, were
used in this investigation. All three compressors utilized
vaneless diffusers and backswept impeller blading,
indications to the dimensions of which are depicted in Table
1. It is worth emphasising that all three geometries are
typical of automotive stages, being devoid of recirculating
casing treatments and pre-swirl vanes. The baseline
geometry (C-4) was designed for an automotive gasoline
engine of 2.0L swept volume.
Table 1: Tested compressor stage geometries
ΔD2 (%) Δb2 (%) Δ(D3/D2) (%)
C-4 - - -
C-5 +30.8 +24.3 -8.33
C-6 +48.7 +25.1 -6.81
2. Modelling The modelling work undertaken for the current study comes
under two headings. Firstly, single passage CFD simulations
to allow the blocked region to be evaluated and the impact
on the flow field to be determined. Secondly, 1-D radial
diffuser modelling, illustrating how the findings from the
CFD study were implemented into a new modelling method.
2.1 CFD Methodology
The chosen package for conducting all CFD simulations
within the current work was ANSYS CFX14.0. In order to
balance the requirements for calculation time and modelling
accuracy, the approach taken was to employ single passage
simulations rather than a full stage calculation. While this
neglects the presence of the scroll volute as is effectively
universally found on turbocharger compressors, it will be
shown that when coupled with the correct post processing
techniques, the results from the single passage simulations
will demonstrate a satisfactory degree of accuracy.
The modelling configuration applied for each of the
compressors followed that as described by Harley et al. [1],
as illustrated in Figure 1. The single passage model was
defined to contain three separate domains, two stationary
domains for the inlet and diffuser respectively, and one
rotating domain for the impeller. As shown in Figure 1, the
total cell count was approximately 1.6 million, a value
arrived at through having conducted a grid independence
study. The Shear Stress Transport (SST) turbulence model
was employed, necessitating y+ to be maintained below five
in all three domains, with a level of less than two being
3
achieved throughout the majority of the model. The
measurement planes (MP) depicted in Figure 1 correspond
with those available from test data and with 1-D interstage
measurement points.
Figure 1: Single passage CFD setup [3]
In terms of solver convergence criteria, convergence was
deemed to have been achieved when the RMS residuals fell
below 1x10-4
[6], the imbalances of mass, energy and
momentum across the model fell below 0.01%, and the
total-to-total isentropic efficiency fluctuations were less than
0.05%. When simulating low mass flow conditions, the
surge point was defined as when the solver failed to meet
the convergence criteria.
In order to better replicate the real compressor stage,
additional loss terms had to be applied to account for aspects
not represented within the single passage models. During the
post processing of the CFD data two additional 1-D losses
were applied, namely the volute loss model of Weber and
Koronowski [7], and the disk friction loss of Whitfield [8].
The resulting CFD predictions are compared with test data
in the next section.
2.2 CFD Validation
In order to establish confidence in the accuracy of the CFD
predictions, the CFD predicted compressor maps of both
total pressure ratio and total-to-total isentropic efficiency
were compared to those gathered during gas stand testing in
QUB. The layout of the test facility is depicted
schematically in Figure 2, with details of the testing
procedure employed described in [3].
Figure 2: Schematic of QUB turbocharger test
facility [3]
It is worth emphasizing that the CFD results presented have
been post processed to include 1-D loss correlations for the
volute and disk friction, as detailed in the previous section.
The resulting comparisons between the single passage CFD
results and test data are presented in Figure 3 to Figure 5 for
C-4 to C-6 respectively.
Figure 3: Comparison of CFD results with test data for C-4
Figure 4: Comparison of CFD results with test data for C-5
4
Figure 5: Comparison of CFD results with test data for C-6
Upon inspecting the results depicted in Figure 3 to Figure 5,
it is apparent that the CFD has predicted performance within
an acceptable window of accuracy. The pressure ratio and
efficiency predictions follow the trend of the test data well,
with the maximum variation between the data sets being
witnessed at low mass flow rates for each of the
compressors. It was demonstrated in previous work ([3])
that the test data is essentially devoid of the effects of heat
transfer, however what little impact it did have on efficiency
would be most prevalent at low speeds and low mass flows.
It is notable that surge is frequently predicted at an
unrealistically low mass flow rate when compared with the
test data, with maximum deviations of 48.4%, 17.2% and
50.7% being witnessed for C-4 to C-6 respectively. It is of
course recognised however that surge is a system
phenomenon that is also fundamentally unsteady; the CFD
model used did not represent the full system and assumed
steady flow since surge prediction was not a primary target
of the modelling work. In addition, for C-4 there were
convergence issues which could not be resolved within the
upper two speed lines, yielding a substantially truncated
map prediction at these operating conditions. These
discrepancies are not entirely unexpected with a single
passage simulation however. While a 1-D volute loss model
has been applied, it is not sufficient to capture the non-
axisymmetric, three-dimensional and unsteady nature of the
volute flow field, and its impact on the performance and
stability of components upstream at off-design conditions
[9].
In fact, the prediction of surge to be at differing flow rates to
the test data does not pose a problem for the current work.
The blockage values used in a subsequent section, which
represent the full extent for which CFD was used in the
current work, were only extracted at mass flow rates that fall
within the bounds of the operating range as defined by the
test data. Therefore, with the magnitude of the CFD
predictions having been shown to be reasonable within the
confines of the test data, the methodology applied is fully
sufficient for the current study.
2.3 Characterisation of Diffuser Blockage
Having gained confidence in the accuracy of the CFD
simulations, it was possible to move forward with the 1-D
modelling work. In order to evaluate the radial extent of this
aerodynamic blockage, the same approach was applied as
was used in the diffuser inlet blockage study referred to in a
previous section, but instead of being evaluated exclusively
at diffuser inlet, it was applied at approximately 25 discrete
radial locations from the inlet to exit of the diffuser
(depending on geometry). Again, the Turbo Chart feature of
ANSYS CFX was utilized to extract circumferentially
averaged streamwise velocity and density values at 250
discrete points equally spaced from hub to shroud at each of
the chosen streamwise locations for each compressor.
Using the continuity equation, it was then possible to
calculate the mass flow passing through each pitchwise
“slice” of the diffuser passage (denoted “dz” in Figure 6).
The extent of the active flow region was then calculated by
summing from hub to shroud until the stage mass flow rate
was reached. The proportion of the remaining diffuser slices
in relation to the total number of 250 defined the extent of
the aerodynamic blockage. A schematic representing this
procedure is depicted in Figure 6.
Figure 6: Determination of diffuser aerodynamic blockage
The result of this analysis is a diffuser passage that is
defined by two separate regions; an active flow region
through which the entirety of the stage mass flow is
assumed to pass, and a blockage region which makes no
contribution to outlet flow conditions. The aerodynamic
blockage region influences the flow field in three ways:
increased radial velocity component VR2 (by
continuity)
reduction in absolute flow angle α2
modification of effective diffuser inlet to outlet
area ratio
The first two parameters directly impact upon the flow field,
while the final one has an indirect impact on actual diffuser
5
performance through modifying the ideal achievable CP for
the diffuser [10], as illustrated in Eq. (3). As will become
apparent in the coming sections, this final element has a
significant impact on diffuser performance.
𝐶𝑃𝐼 = 𝑐𝑜𝑠2(𝛼2) (1 −1
𝐴𝑅2)
+ 𝑠𝑖𝑛2(𝛼2) (−1
𝑅𝑅2)
(3)
In order to capture the variation of blockage across the
different geometries and operating conditions being tested, it
was deemed necessary to formulate a simplified method
reliant on data from only a number of operating points,
rather than the full compressor map. The variation of the
blockage throughout the diffuser was extracted at surge,
choke and peak efficiency for all three geometries at a
number of operating conditions. A sample result from this
analysis is presented in Figure 7 for C-6 at 75% speed,
detailing the CFD results as well as the correlations
generated to replicate the blockage variation on a 1-D basis.
Figure 7: Diffuser blockage variation for C-6 at 75% speed
The trend illustrated in Figure 7 is one that was echoed
across all three geometries at a range of operating
conditions; the high levels of inlet blockage at surge
depicted a tendency to decrease through the diffuser, while
the comparatively low levels of inlet blockage at choke
tended to increase with radius in the diffuser. At the surge
side of the map, low velocity fluid is subject to a strong
adverse pressure gradient, aggravating the tendency for
boundary layer separation and recirculation. The
relationship between blockage and diffuser radius ratio
witnessed in Figure 7 can be attributed to the decay of this
recirculation zone present close to impeller exit, as depicted
in Figure 6. Towards choke, the fluid velocity is greater and
the adverse pressure gradient much less severe. As a result,
the recirculation region which is dominant at surge is no
longer present, meaning the blockage presented to the flow
is predominantly attributable to boundary layer growth from
inlet to outlet of the diffuser. Therefore, having analysed the
blockage levels throughout the diffuser for the three
geometries under consideration at a range of operating
conditions, it became apparent that the overall trend could
be well represented (albeit on a simplified basis) at both
surge and choke by exponential functions, as illustrated in
Figure 7.
The proposed correlations, as illustrated in Eq. (4) and (5),
represent either an exponential decay or growth of the
diffuser aerodynamic blockage for surge and choke
respectively. The exponent “x” represents radial location in
the diffuser, which equates to the quotient of the current
calculation step, i, and the total number of calculation steps.
Throughout the current analysis performance was evaluated
using 100 radial steps through the diffuser, a value that was
deemed to balance the competing aims of calculation time
and prediction accuracy.
𝐵𝑖,𝑠𝑢𝑟𝑔𝑒 = 𝐵2𝑒−2𝑥 (4)
𝐵𝑖,𝑐ℎ𝑜𝑘𝑒 = 𝐵2𝑒0.5𝑥 (5)
At this stage, representative correlations have been
generated for surge and choke, however no consideration
has been given for mid map conditions. In order to
accurately represent the blockage present away from the
extremities of map width, a linear interpolation procedure
was employed on the basis of the diffuser inlet flow angle
α2, as depicted in Eq. (6). This approach provided a smooth
transition between the two correlations and, as will become
apparent in the coming sections, a good estimation of
diffuser performance across the operating range without the
need for extensive knowledge about the flow field at every
radial location in the diffuser at each operating point.
𝐵𝑖
= 𝐵𝑖,𝑠𝑢𝑟𝑔𝑒 + (𝐵𝑖,𝑐ℎ𝑜𝑘𝑒
− 𝐵𝑖,𝑠𝑢𝑟𝑔𝑒) (𝛼2 − 𝛼2,𝑠𝑢𝑟𝑔𝑒
𝛼2,𝑐ℎ𝑜𝑘𝑒 − 𝛼2,𝑠𝑢𝑟𝑔𝑒
)
(6)
3. 1-D Diffuser Modelling Based on the findings from the previous study [3], it was
deemed that the most appropriate model to build upon for
the current work was a model based upon an equation first
presented by Rodgers [11], but subsequently converted into
a vaneless diffuser modelling method by Stuart et al. [3].
The Rodgers equation permitted evaluation of diffuser CP
directly, knowing only the overall diffuser geometry and
inlet flow angle. This was further developed to calculate
diffuser pressure recovery for each radial step, as well as the
associated key flow parameters using a compressible
analysis.
The reasoning behind this was that, ultimately, the aim was
to develop a vaneless diffuser model for which each of the
individual sources of loss were accounted for, and with the
simplistic method outlined above only currently accounting
for the impact of wall friction, it was deemed a good basis to
build upon. In order to evaluate the improvement in
prediction over existing methods, the Herbert model [5],
which is an extension and correction of Stanitz’ [12] ground
breaking work, was chosen as it represented the most
6
complex of the existing approaches, incorporating terms for
boundary layer growth and total pressure loss as a function
of Mach number and boundary layer shape factor.
Fundamentally, the proposed model evaluates the coefficient
of static pressure rise, CP, for each calculation step in the
diffuser, based upon geometry and a chosen value of skin
friction coefficient Cf using the aforementioned Rodgers
equation [11], as depicted in Eq. (7). Unlike a number of
existing approaches which effectively employ the skin
friction coefficient as a bulk loss term, it is intended that that
the variable in the proposed model accounts only for wall
friction loss.
𝐶𝑃 = [1 − (𝐷(𝑖)
𝐷(𝑖+1)
)
2
]
−𝐶𝑓
cos (𝛼𝑖)
𝐷(𝑖)
𝑏(𝑖)
(1 −𝐷(𝑖)
𝐷(𝑖+1)
) (7)
From this, the ideal isentropic coefficient of static pressure
rise, CPI, is evaluated using Eq. (3), and compared with the
actual value calculated using Eq. (7) to allow the vaneless
diffuser loss coefficient YD to be calculated, as shown in Eq.
(8).
𝑌𝐷 = 𝐶𝑃𝐼 − 𝐶𝑃 (8)
Knowing the value of YD, it is possible to calculate the
change in total pressure across the current calculation step,
and hence evaluate the total pressure applicable to the
following step, as illustrated in Eq. (9) and Eq. (10)
respectively.
𝑑𝑝𝑜 = 𝑌𝐷(𝑝𝑜(𝑖) − 𝑝(𝑖)) (9)
𝑝𝑜(𝑖+1) = 𝑝𝑜(𝑖) − 𝑑𝑝𝑜 (10)
Before moving on to calculate the remaining flow variables,
it is necessary to incorporate the impact of the previously
described aerodynamic blockage on the flow field. In order
to incorporate the blockage correlations into a 1-D
modelling method, the direct impact on the flow field had to
be evaluated. The first parameters requiring evaluation for
each calculation step were the diffuser inlet flow parameter
ϟ, and the associated quartic relationship for diffuser inlet
blockage B2, as depicted in Eq. (1) and (2) respectively.
Knowing these parameters, and the exponent x, it is possible
to evaluate the surge and choke blockage relationships given
by Eq. (4) and (5), and ultimately calculate the blockage at
any step using the linear interpolation based on absolute
diffuser inlet flow angle α2 detailed in Eq. (6).
Knowing the blockage presented to the flow, it was possible
to evaluate the direct impact on the radial velocity using the
continuity equation, as illustrated in Eq. (11).
𝑉𝑟𝑖 =��
𝜌𝑖(1 − 𝐵𝑖)(2𝜋𝑟𝑖𝑏𝑖) (11)
As the analysis is compressible in nature, a further internal
iterative process was incorporated into the analysis to
determine the density at each step. This involved
undertaking the above procedure, then continuing the
calculation to determine the total and static pressure and
temperature values associated with the newly calculated
flow conditions. From the resulting static pressure and
temperature values, a new density was calculated using the
equation of state, and compared with the initial value. If
satisfactory convergence between the two density values
was not achieved, the entire calculation was completed
again, beginning with the determination of the blockage
level, until such a point where the solution was deemed to
have converged.
Having specified the basis of the proposed modelling
method, it was possible to evaluate its impact on the
prediction of diffuser performance. The parameter upon
which the modelling methods will be evaluated is the
coefficient of static pressure rise (CP). CP, as depicted in
Eq. (12), captures the fundamental purpose of a centrifugal
compressor diffuser, by providing a metric to determine the
proportion of total pressure at inlet of the diffuser that was
successfully converted into static pressure at the exit of the
diffuser.
𝐶𝑃 =𝑝3 − 𝑝2
𝑝02 − 𝑝2
(12)
As a first step, a direct comparison of diffuser CP prediction
was undertaken for each of the geometries. This utilised the
same methodology as was applied by Stuart et al. [3], where
diffuser inlet conditions were specified entirely from testing
data to ensure the integrity of the comparison and to remove
any discrepancies attributed to limitations in the meanline
impeller modelling. The resulting comparison between test
data, the Herbert model and the proposed model for each of
the three geometries is presented in Figure 8.
In order to improve the clarity of the resulting comparison,
only three speedlines (representing low, medium and high
tip speeds) for each geometry are presented. Furthermore,
the values of CP in each case have been non-
dimensionalised using the maximum value of CP achieved
across all three geometries, be that from test results or 1-D
predictions. It must be noted that this approach is in
contravention of what was conducted for the other figures
within the manuscript, where the maximum value for a
given parameter was taken for each compressor
individually. However, it helps to illustrate a key
observation within the discussion that follows.
7
Figure 8: 1-D diffuser modelling comparison
Discussion What is immediately apparent from analysing Figure 8 is
that the baseline Herbert model is lacking the ability to
predict the trend of how CP varies with α2 across the
compressor map. Despite the relative complexity of the
modelling method, as explained in the preceding sections,
the predominant feature of the approach is the impact of the
skin friction coefficient Cf. This dictates that CP variation
across the map will be mainly reliant on the relationship
between the chosen friction coefficient, and the flow path
length within the diffuser. As a result, for a given value of
Cf, CP will be maximum for the lowest value of α2, which
corresponds to the shortest flow path, and consistently
decrease as the flow angle becomes more tangential. Due to
the additional parameters in the model relating the loss in
total pressure through the diffuser to the boundary layer
shape factor, which in turn is dependent upon Mach number,
there is some spreading of the speedlines at a constant value
of α2. However, even with this addition the influence of the
friction loss is still predominant.
A further point worth noting is that it is evident that,
generally speaking, the 1-D modelling methods deliver
better correlation with the test data for C-5 and C-6 than for
C-4, which has the smallest impeller but the largest diffuser
radius ratio of the three test cases. It is readily apparent from
Figure 8 therefore that the increase in the ideal CP, as
depicted in Eq. (3), brought about by the larger radius ratio
(RR) of C-4 is not sufficiently counteracted by the increased
frictional losses associated with the longer flow path for a
given increase in flow angle within the diffuser.
With respect to the proposed modelling method, it is clear
that it offers an improved prediction, particularly at the
extremes of the compressor operating map. At higher values
of α2 the approach of directly applying the loss in total
pressure associated with the skin friction coefficient in each
geometry step delivers a more accurate representation of
real performance when compared with the Herbert model,
illustrated by the substantial performance decrement
towards surge. For example, considering the maximum α2
values reached by C-5 for each of the three speedlines under
consideration, the maximum deviation from the test data
witnessed with the proposed model was 1.0% compared
with 25.7% for the Herbert model. Similarly, for C-6 the
largest discrepancy with the test data witnessed was 13.6%
compared to a value of 43.6% with the Herbert model. As
described above, the correlation achieved for C-4 was not as
impressive as for the other two geometries, however the
proposed method still delivers a more representative
prediction than that achieved by the Herbert model.
At the opposite side of the performance map, the
performance decrement towards choke that is evident in the
test data, which is absent in the Herbert prediction, is
captured well by the proposed model, especially for C-5 and
C-6. Comparing again the maximum deviation between the
test data and each of the modelling methods, this time at
minimum α2, for C-5 this amounted to 7.0% for the
proposed model, compared to 22.2% for the Herbert model.
Similarly, for C-6 the maximum deviation of 12.9% for the
proposed model compares favourably with the 48.6%
arising from the Herbert model. In the same vein as
previously mentioned, the prediction from the proposed
method for C-4 did not follow as well as for the other two
geometries, but still offered an improvement over the
baseline Herbert method. While the reasoning behind this
has been covered in the preceding paragraphs, further work
would be required to gain an understanding of the best way
to modify the model to account for the difference in
performance for C-4.
In terms of an explanation for the drop off in performance
towards choke (low values of α2) within the proposed
model, the blockage correlation for choke as depicted in Eq.
(5) dictates that from inlet to outlet, the level of blockage
presented to the flow must increase. As a result, the
8
effective geometry of the diffuser has been modified,
meaning the area increase associated with each calculation
step is much less than would be expected taking only the
physical geometry into account. A simplified schematic
illustrating this point are depicted in Figure 9.
Figure 9: Schematic of diffuser geometry with blockage
As is illustrated in Figure 9, while the levels of blockage at
the inlet of the diffuser are much smaller at choke than at
surge, it is the growth of the blocked region towards choke
that causes the significant decrement in performance. As the
effective passage height is reducing for each calculation
step, the area increase associated with the increase in radius
is diminished over the case with no blockage, or the surge
case. Going back to Eq. (3), the reduction in the area ratio of
the diffuser brings about a reduction in the ideal achievable
CP, resulting in the ability to replicate the performance
decrement witnessed in the test data. It is clear therefore
from the current model that wall friction is still the
predominant 1-D source of loss towards surge, but
aerodynamic blockage throughout the diffuser has been
identified to be a significant contributing factor towards the
choke side of the map.
3.1 Impact on 1-D Stage Calculation
As a final verification of the benefit of the proposed
vaneless diffuser model, the decision was taken to evaluate
the impact on a 1-D stage performance prediction,
incorporating models for the impeller, vaneless diffuser and
volute. The impeller losses applied to the model originated
from the work of Galvas [13], with the inlet recirculation
model of Harley et al. [1] and the choking loss of Aungier
[14] was also applied. The slip factor correlation applied
was that of Qiu et al. [15], with the volute loss again being
accounted for using the work of Weber and Koronowski [7].
As with the previous section detailing a diffuser only
performance evaluation, the vaneless diffuser model used
for comparison against the proposed model was that of
Herbert [5].
The resulting comparisons for C-4 to C-6 are illustrated in
Figure 10 to Figure 12 respectively, where all data have
been normalised using the maximum respective value for
that parameter. As with the previous section, to preserve the
clarity of the comparison only three speed lines per
geometry have been presented, representing low, medium
and high tip speed operation.
Figure 10: Comparison of 1-D modelling with test data for
C-4
Figure 11: Comparison of 1-D modelling with test data for
C-5
CHOKE SURGE
9
Figure 12: Comparison of 1-D modelling with test data for
C-6
Discussion
What is instantly striking about the comparisons illustrated
in Figure 10 to Figure 12 is that the magnitude of the
improvement in the prediction achieved on a diffuser only
basis has not been reflected as significantly on a stage
performance basis. Generally speaking, the change in
performance associated with the proposed model has a
greater impact on the choke side of the map, however while
this is advantageous in bringing the prediction closer to the
test data at higher speeds, it generally has a detrimental
impact on the lower speed lines. This observation is
particularly pertinent for C-6, as depicted in Figure 12.
A common observation for each of the geometries is that the
accuracy of both the pressure ratio and efficiency
predictions diverge as compressor speed increases. The
trend depicted in the test data concerning the drop in peak
compressor efficiency at high speeds is one that is not
captured by the existing 1-D modelling technique, resulting
in diminishing efficiency prediction with increasing speed.
As noted by Harley et al. [1], this trend is common in
automotive turbocharger compressors, where designs are
specifically targeting improved efficiency at low speeds and
mass flow rates to better align with the engine transients
encountered during urban driving.
What is also evident is while the efficiency prediction
diverges at higher speeds, the prediction of pressure ratio
does so at an increased rate. Taking C-5 as an example, on
the 100% speedline at m/mmax of 0.75, the error in the
efficiency prediction equates to 5.6% between the test data
and proposed model, while the pressure ratio prediction
illustrated an associated error of 21.3%. Upon investigating
this problem more deeply, it became apparent that the
discrepancy was related to the prediction of the tangential
velocity at impeller exit (Vu2) from the single zone model.
Comparing the prediction of Vu2 from the single zone model
to that from the previously described CFD simulations, an
over prediction of 19.2% by the single zone model was
evident for the same operating point for C-5. It would
appear that the issue here relates to the determination of the
impeller slip factor, as if it was over predicted, it would
manifest itself as an increase in predicted pressure ratio (in
accordance with the Euler equation). However, as slip factor
is not a loss, it would not have the same impact on
efficiency, which is exactly the symptom depicted in the
current data set.
A further issue with the 1-D prediction is the prediction of
the choking mass flow rate, the accuracy of which again
diverges with increasing compressor speed. The current
method applied is that of Dixon & Hall [16], which
compares the available inducer throat geometric area to a
calculated choking area based on total inlet conditions.
Unfortunately, this simplified approach cannot account for
the highly non-uniform velocity profile at the throat section
and the associated boundary layer blockage [17], which like
in the current work, results in the full geometric area not
being available to pass the stage mass flow. Consequently,
choking in the real stage occurs significantly earlier than
what is predicted by the simplistic 1-D analysis currently
applied. Again, further work is required to improve the
fidelity of the choking model to account for these real flow
effects.
4. Conclusions Improvements in the 1-D prediction of vaneless diffuser
performance have been achieved through the analysis of
three automotive turbocharger centrifugal compressor
stages. A combination of extensive gas stand test data,
which was shown to have been gathered under
approximately adiabatic conditions, as well as single
passage CFD simulations for each geometry were employed
to permit full characterisation of the diffuser flow field.
Modelling of the extent of the aerodynamic blockage
presented to the flow throughout the diffuser with changing
geometry and operating conditions was completed on a
simplified basis, and was illustrated to have a significant
impact on diffuser performance. Correlations describing the
variation in the aerodynamic blockage throughout the
diffuser resulting from the stratified flow field emanating
from the impeller were developed, and incorporated into a
new diffuser modelling method described herein.
Utilising the methodology developed by the authors for a
previous study, direct comparisons were drawn between the
diffuser performance prediction delivered by the proposed
model, the baseline 1-D Herbert model and test data. By
providing the 1-D models and test data with effectively
common input parameters, a robust diffuser-only
performance analysis was conducted for each of the
geometries, validating significant improvements in the off-
10
design performance prediction delivered by the proposed
model. This improved performance prediction for vaneless
diffusers is certainly beneficial at the preliminary design
stage, allowing the designer to consider appropriate
geometry before committing to 3D CFD modelling.
Upon incorporating the proposed diffuser model into a 1-D
stage calculation, the limitations in the modelling of the
other stage elements (particularly the impeller) masked the
benefits witnessed on a diffuser-only basis. Comparing the
1-D stage prediction with test data for each geometry did
however provide some guidance for future work, with the
divergence in pressure ratio and efficiency prediction
towards higher speeds and the overall choking mass flow
prediction providing focal points.
Acknowledgements The authors would like to sincerely thank IHI Charging
Systems International GmbH for provision of the necessary
compressor hardware for testing, as well as for their
continuing technical support. The authors would also like to
extend their thanks to ANSYS Inc. for the use of their CFD
software and their technical support during this programme
of research.
References
[1] Harley, P., Spence, SW., Filsinger, D., Dietrich, M., &
Early, J., 2015, “Meanline Modeling of Inlet
Recirculation in Automotive Turbocharger Centrifugal
Compressors,” ASME Journal of
Turbomachinery,Vol.137 (1)
[2] Qiu, X., Japikse, D., & Anderson, M., 2008, “A
Meanline Model for Impeller Flow Recirculation,”
Proceedings of ASME Turbo Expo 2008, Berlin,
Germany, GT2008-51349
[3] Stuart, C., Spence, S.W., Filsinger, D., Starke, A.,
Kim, S., & Harley, P., 2015, “An Evaluation of
Vaneless Diffuser Modelling Methods as a Means of
Improving the Off-Design Performance Prediction of
Centrifugal Compressors,” Proceedings of ASME
Turbo Expo 2015, Montréal, Canada, GT2015-42657
[4] Cumpsty, N.A., 2004, “Compressor Aerodynamics,”
Krieger Publishing Company, Florida, ISBN 1-57524-
247-8, pp277
[5] Herbert, M.V., 1980, “A Method of Performance
Prediction for Centrifugal Compressors,” ARC R&M
No. 3843, H.M. Stationery Office, London, England
[6] Stern, F., Wilson, R.V., Coleman, E.W., & Paterson,
E.G., 2001, “Comprehensive Approach to Verification
and Validation of CFD Simulations – Part 1:
Methodology and Procedures,” ASME Journal of
Fluids Engineering, Vol. 123 (4)
[7] Weber, C.R., & Koronowski, M.E., 1986, “Meanline
Performance Prediction of Volutes in Centrifugal
Compressors,” Proceedings of ASME Turbo Expo,
Dusseldorf, Germany, 86-GT-216
[8] Whitfield, A., 1974, “Slip Factor of a Centrifugal
Compressor and its Variation with Flow Rate,” Proc
Instn Mech Engrs Vol. 188 32/74 pp415-421
[9] Yang , M., Zheng, X., Zhang, Y., Bamba, T., Tamaki,
H., Huenteler, J., & Li, Z., 2010, “Stability
Improvement of High-Pressure-Ratio Turbocharger
Compressor by Asymmetric flow Control: Part I-Non
Axisymmetric flow in Centrifugal Compressor,”
Proceedings of ASME Turbo Expo 2010, Glasgow,
Scotland, GT2010-22581
[10] Whitfield, A., & Baines, N.C., 1990, “Design of
Radial Turbomachinery,” Longman Scientific &
Technical, Essex, England, pp118
[11] Rodgers, C., 1984, “Static pressure recovery
characteristics of some radial vaneless diffusers,”
Canadian Aeronautics and Space Journal, Vol. 30,
Issue, 1, pp42-54
[12] Stanitz, J.D., 1952, “One-dimensional compressible
flow in vaneless diffusers of radial- and mixed-flow
centrifugal compressors, including the effects of
friction, heat transfer and area change,” National
Advisory Committee for Aeronautics, Technical Note
2610, Washington, USA
[13] Galvas, M. R., 1974, “Fortran Program for Predicting
Off-Design Performance of Centrifugal Compressors,”
NASA Lewis Research Center, Cleveland, OH, Paper
No. NASA-TN-D-7487.
[14] Aungier, R. H., 2000, “Centrifugal Compressors: A
Strategy for Aerodynamic Design and Analysis”,
ASME Press, New York.
[15] Qiu, X., Mallikarachchi, C., and Anderson, M., 2007,
“A New Slip Factor for Axial and Radial Impellers,”
Proceedings of ASME Turbo Expo 2007, Montréal,
Canada, GT2007-27064
[16] Dixon, S.L., & Hall, C.A., 2010, “Fluid Mechanics
and Thermodynamics of Turbomachinery,”
Butterworth-Heinemann, USA, pg 256-258
[17] Van den Braembussche, R,A,, 2014, “Centrifugal
Compressors Analysis & Design,” VKI Course Note
192, pg 72